Introduction
Arterial spin labelling (ASL) is a magnetic resonance imaging (MRI) technique increasingly used in research and clinical settings to calculate cerebral blood flow (CBF) non-invasively [
1]. It has been recognised that ASL can provide an early biomarker for dementia, cognitive decline and small vessel disease [
2‐
8]. Despite statistical differences between groups, e.g. Alzheimer’s disease, mild cognitive impairment and normal ageing, clinical applications have been hampered by unexplained inter-subject variability [
9,
10].
Deriving quantitative perfusion values from the raw MRI signal requires the application of a model containing several assumptions that relate to physiological properties of the blood and tissues. The white paper recommendations of the International Society for Magnetic Resonance in Medicine (ISMRM) and the European Consortium for ASL in Dementia propose pseudo-continuous ASL (PCASL) with a single post-labelling delay (PLD) and slice-timing correction, and the application of a simplified Buxton equation for quantification of CBF [
11,
12]. In this model, 1650 ms is recommended as the longitudinal relaxation time of blood (T1
blood) at 3 T. This has been derived from a linear relationship between haematocrit (Hct) and blood T1 found in experiments under appropriate physiological conditions [
13], assuming an average adult Hct of 43.5% [
11]. However, there are well-recognised sources of variation in Hct between and within populations that may render CBF estimation inaccurate if they are excluded from the model [
9,
10].
Hct varies by sex with females typically having lower Hct than males [
14]. Most studies show a lower Hct in blacks than Asians compared with whites [
15,
16], although an earlier study found this varied according to age and was not applicable in some younger subjects (men aged 15–24) [
17]. Higher Hct levels have been associated with obesity [
14] and have also been associated with risk of developing diabetes [
18]. Diabetic patients with long-standing disease may have decreased Hct, possibly due to diabetic nephropathy causing erythropoietin deficiency [
19], or malabsorption of vitamin B
12 as a side-effect of long-term treatment with metformin [
20].
The purpose of this study was to identify the variability in cortical CBF using ASL, to investigate the influence of Hct on the estimation of CBF and to determine how this impacts on the associations of CBF with age, sex, ethnicity and diabetes.
Materials and methods
Study population
Subjects (
n = 493, 40% female, age range 55–90) were recruited from the Southall and Brent Revisited (SABRE) study. Approval for investigations was obtained from the Fulham Research Ethics Committee (ref: 14/LO/0108) and participants gave written informed consent. SABRE is a longitudinal study principally investigating cardio-metabolic disease in a tri-ethnic population cohort; details have been published elsewhere [
21]. Briefly, participants were community-dwelling elderly men and women and resident in the north and north-west London at the commencement of the SABRE study in 1988. Ethnicity was defined based on the country of origin of all four grandparents. Spouses or significant others were invited to participate in a clinical visit, commencing in 2014. An additional booster sample of African Caribbeans from the same area of London was recruited to increase numbers and therefore enhance the power of the study. Data for this study were acquired between August 2014 and October 2016.
Investigations
Participants underwent all examinations on the same day as MRI. Participants were requested to consume no more than an early light breakfast prior to arrival at the clinic. Weight and height were measured using a Tanita Pro BC-418 Body Composition Analyser and a Seca 216 Stadiometer, respectively, and body mass index (BMI) was calculated. Supine central blood pressure (BP) and heart rate (HR) were measured using a Pulsecor device after 5-min rest, in both arms. BP was calculated as the average of the final two of three measurements in the left arm unless the difference between arms was > 10 mmHg in which case the arm with the higher BP was used. Mean arterial pressure (MAP) was calculated using a Pulsecor device (
www.uscom.com.au) to measure the arterial pressure waveform during suprasystolic brachial cuff inflations. Hct was measured using an impedance-based, direct current sheath flow method (Sysmex XE-2100) from a venous blood sample drawn on the same morning as the MRI examination. Blood was also analysed for glycosylated haemoglobin (HbA1c), serum total cholesterol, high-density lipoprotein (HDL) cholesterol and triglycerides. Low-density lipoprotein (LDL) cholesterol was calculated using the Friedewald equation.
Diabetes mellitus
Diabetes status was determined by the following criteria: primary care records of diagnosis or prescription of diabetes medication, patient recall of diagnosis or taking antidiabetic medication, fasting or oral glucose tolerance test or plasma glucose testing from previous SABRE study visits using the WHO recommendations [
22] and those with an HbA1c > 47 mmol/mol when tested in the research clinic.
MRI imaging protocol
All subjects were examined at a single centre, University College Hospital, on a 3 T MRI (Achieva, Philips Healthcare) using an 8-channel phased-array head coil. The protocol included a sagittal T1-weighted 3D-TFE (TR/TE/TI 7/ 3.2/836 ms, flip-angle 18°, voxel size 1 mm3) and a transversal 2D pseudo-continuous arterial spin labelling (PCASL) (EPI, TR/TE 4615/15 ms, flip-angle 90°, voxel size 3.75 mm × 3.75 mm × 5 mm, 1-mm slice gap, 20 slices), labelling duration 1800 ms, post-labelling delay 2000 ms, 35 acquisitions, 2 background suppression pulses (1950 ms and 3296 ms after saturation). There were three repetitions of a proton density–weighted image with TR = 9000 ms and no background suppression, but otherwise identical parameters as PCASL were also acquired. Planning was aligned to the anterior commissure-posterior commissure line in the transversal plane orthogonal to the T1w sagittal plane, ensuring coverage of the entire cerebrum including the vertex. The labelling plane was positioned on a phase-contrast survey to identify the vessels.
MR segmentation and cerebral blood flow processing
Tissue segmentation and region labels were obtained using the Geodesic Information Flows framework [
23] (
https://github.com/KCL-BMEIS/NiftySeg). This method produces a state-of-the-art segmentation and regional labelling by voxel-wise voting between several propagated atlases, guided by the local image similarity. Grey matter and white matter are defined within the propagated atlases. Segmentations of the grey matter, white matter and cerebrospinal fluid space are resampled to the space of the ASL acquisition making use of the known point-spread function to account for down-sampling induced loss of information.
The determination of CBF maps from ASL data followed the simple derived form for PCASL (Eq.
1) from [
11] presented in units of mL/100 g/min using an open-source in-house software package [
24] (
https://cmiclab.cs.ucl.ac.uk/CMIC/NiftyFit-Release). Thirty-five control, (SC) and label (SL) pairs and 5 proton density (SPD) images were averaged to generate single voxel values for the control and label in Eq.
1, (where λ is the blood/brain partition coefficient (0.9 mL/g), PLD is the post-labelling delay between the end of bolus and the start of imaging (2000 ms), T1
blood is the blood T1 relaxation time, α is the labelling efficiency (85%) and τ is the labelling duration (1800 ms). T1
blood was calculated based on the formula: T1 = (0.52 × Hct + 0.38)
−1[
13] either with fixed value of 43.5% (corresponding to the standard value of T1 = 1650 ms), which was used in model 1 (CBF
fixed), or calculated based on the Hct values measured from each participant and used in model 2 (CBF
Hct). The perfusion difference between CBF models using a fixed Hct and individualised Hct was calculated as an absolute number and as a percentage of the fixed Hct model.
$$ \mathrm{CBF}=6000\uplambda\ {\mathrm{e}}^{\mathrm{PLD}/\mathrm{T}1\mathrm{blood}}\ \left(\mathrm{SC}\hbox{--} \mathrm{SL}\right)/\mathrm{SPD}\ \left(2\upalpha\ {\mathrm{T}1}_{\mathrm{blood}}\left(1-{\mathrm{e}}^{-\uptau /\mathrm{T}1\mathrm{blood}}\right)\right)\left[\mathrm{mL}/100\ \mathrm{g}/\min \right] $$
(1)
Partial volume correction (PVC) was applied based upon the method used in [
25]. Results for segmented cortical grey matter PVC CBF are presented. Cortical grey matter CBF without PVC are presented in supplemental material (Supp Table
1).
Statistical analysis
Analysis was performed using STATA 14.2 (College Station, TX). All analyses were stratified by sex and ethnicity. Continuous data are presented as mean and standard deviation (SD); categorical data are counts and percentages.
Paired t tests were used to test for statistical significance between the two methods of CBF calculation. Pearson’s correlation coefficient was computed to quantify the correlations between Hct and CBF estimates. Comparisons by sex and ethnicity were performed using two-way analysis of variance (ANOVA), followed by individual group-wise comparisons using Fisher’s least significant difference test if ANOVA was significant (p < 0.05). Multiple linear regression analyses were performed to assess participant characteristic associations with Hct levels with further adjustment for potential confounders, age, mean arterial pressure, diabetes, HbA1C, LDL cholesterol, HDL cholesterol and BMI, chosen a priori. Sample frequency distributions of CBF calculated with and without correction for individual Hct were analysed using univariate kernel density estimates.
Discussion
This study has shown that Hct levels differ according to sex and ethnicity and that this influences CBF estimated from ASL. Importantly, failure to adjust T1
blood according to sex and ethnic variation in Hct leads to a significant overestimation of CBF in women and non-European populations. Although other studies have investigated the effect of Hct on CBF estimation with ASL in sickle cell [
26] and neonatal [
27] groups; this is to the best of our knowledge, the first study to address this in a community-based elderly multi-ethnic population.
A study by Parkes et al [
28] using continuous ASL found females had higher grey matter CBF than males by 13%. It seems likely that failure to measure individual Hct influenced these observations. We found a similar difference in CBF between white European men and women (+ 9.1% for women) when neither Hct nor PVC was accounted for, but this difference was completely abolished by correction for Hct and PVC. Following correction for PVC and Hct, CBF was slightly lower for women than for men in South Asian (− 5.6%) and African Caribbean (− 6.0%) ethnicities. This finding is broadly aligned with an
15O PET study where gender differences in CBF evident in younger subjects were not significant in subjects older than 65 years [
29]. It has been suggested that gender differences in young subjects are due to high oestrogen levels in females causing CBF fluctuations during the menstrual cycle [
30]. Our large elderly sample scanned with age-appropriate parameters such as a long PLD and with CBF calculated using individualised values is likely to provide a more accurate reflection of true physiological differences than studies that have used a ‘one size fits all’ protocol.
We also demonstrated that the utilisation of individually measured Hct in the calculation of T1
blood reduced the inverse correlation between CBF and Hct. However, some association remained in men even after adjustment for mean arterial blood pressure, diabetes and dyslipidemia. Increased CBF is an expected response to decreases in haemoglobin (and thus Hct) as a mechanism to sustain the cerebral metabolic rate of oxygen levels (CMRO
2) as demonstrated by Ibaraki et al [
31]. It has also been suggested that higher Hct levels and concomitant increases in blood viscosity may influence capillary flow due to alterations in functional shunting [
32]. The higher Hct levels found in men may provide an explanation why we found some association between Hct and CBF only in men as increasing Hct increases blood viscosity in a non-linear relationship, and it has been suggested that increased viscosity decreases CBF [
33].
Lower values of Hct in those with diabetes indicate that further investigation using larger samples is warranted to investigate the interactions of sex, ethnicity, Hct and the effect of diabetes on CBF.
Investigation of the association of age and CBF, which had yielded conflicting evidence in previous studies [
10,
28,
34], was not the main aim of our study. However, it is noteworthy that we found an association between age and CBF in the non-partial volume-corrected data which disappeared following partial volume correction. A likely explanation for this is the decreased brain volume in older subjects which makes the ASL perfusion data more prone to partial volume effects with non-perfused CSF, thereby leading to an artificial CBF decrease. However, PVC may introduce overestimation of cortical CBF particularly in examinations hampered by head movement or poor SNR, either in the T1 structural images used for segmentation or in the ASL acquisition which as a subtraction technique is particularly susceptible to movement artefact. Some movement inevitably remains in a large population study of elderly individuals and this may be a potential source of under- or overestimation of partial volume-corrected CBF in our cohort.
Despite a more diverse ethnic sample, our cortical grey matter CBF values are comparable with previous studies using PCASL on elderly, community-dwelling populations [
6,
7]. PCASL has some limitations when applied to elderly subjects. One factor affecting CBF quantification is blood velocity relative to the labelling plane. Any discrepancies from the expected range of blood velocities due to vascular pathologies such as internal carotid stenosis or vessel tortuosity might result in reduced labelling efficiency and therefore CBF underestimation. Our study used 2D PCASL, rather than a 3D technique, which may be liable to decreased SNR due to longer T1 blood relaxation in slices near the vertex due to sequential slice acquisition [
35], despite using slice-timing correction. Inefficiency of background suppression pulses during multiple slice acquisitions may also have affected the ASL signal.
The main strength of this study was its large community-based, elderly, ethnically diverse sample studied in a single centre so that all protocols were consistently applied throughout. Limitations of the study include a potential selection bias towards healthy individuals who were willing and able to attend the research clinic. Another limitation was the absence of direct T1
blood measurements, although this was mitigated by the use of a previously published model to account for the effect of Hct [
13]. Hct is the main determinant of T1
blood, but other factors such as serum ferritin and HDL cholesterol may also contribute to a minor extent [
36]. Measurement of arterial rather than venous Hct would have been more appropriate to estimate CBF. However, measuring arterial Hct in clinic was impractical and it was considered that venous and arterial Hct were closely related and importantly that venous Hct could be sampled consistently [
37]. Although T1
blood can be easily directly measured in the left ventricle and has lower test-retest variability compared with measured Hct [
38], comparable sequences in the brain are more challenging due to partial volume effects, blood velocity and pulsatility in the measured vessel [
27,
39].
Our findings suggest that research studies using ASL to measure CBF should routinely measure Hct to adjust T1blood, especially in inhomogenous samples. Alternatively, substitution of an appropriate gender- and ethnicity-specific Hct value derived from population group means or direct measurement of T1blood during MRI would improve accuracy of CBF measurement. Further research is warranted into whether adjustment to the Hct value in CBF models to accommodate demographic and pathological differences provides stronger associations with cerebrovascular disease, dementia and cognitive decline than previous models using a fixed mean Hct value. Results from previous studies may need to be interpreted with caution where there are ethnic, gender and pathological differences in the sample. Such an approach may improve early risk assessment in ethnic groups and identify potentially vulnerable groups such as those with known vascular or metabolic disease.
In conclusion, we demonstrated that CBF values obtained from ASL using a fixed Hct mean may lead to systematic errors, resulting most frequently in an overestimation of CBF in female subjects and non-Caucasian ethnicities. It is important to be aware of this when using CBF threshold values to assess disease status or severity. This applies not only to the discrimination between normal ageing and a neurogenerative disease such as AD but could potentially also influence the discrimination between high- and low-grade brain tumours or determination of penumbral threshold values in stroke. We therefore argue that whenever possible, individualised measures of Hct should be included in CBF calculations by ASL.
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.